2 * Copyright (C) 2008 Red Hat. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/pagemap.h>
20 #include <linux/sched.h>
21 #include <linux/slab.h>
22 #include <linux/math64.h>
23 #include <linux/ratelimit.h>
25 #include "free-space-cache.h"
26 #include "transaction.h"
28 #include "extent_io.h"
29 #include "inode-map.h"
31 #define BITS_PER_BITMAP (PAGE_CACHE_SIZE * 8)
32 #define MAX_CACHE_BYTES_PER_GIG (32 * 1024)
34 static int link_free_space(struct btrfs_free_space_ctl *ctl,
35 struct btrfs_free_space *info);
36 static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
37 struct btrfs_free_space *info);
39 static struct inode *__lookup_free_space_inode(struct btrfs_root *root,
40 struct btrfs_path *path,
44 struct btrfs_key location;
45 struct btrfs_disk_key disk_key;
46 struct btrfs_free_space_header *header;
47 struct extent_buffer *leaf;
48 struct inode *inode = NULL;
51 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
55 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
59 btrfs_release_path(path);
60 return ERR_PTR(-ENOENT);
63 leaf = path->nodes[0];
64 header = btrfs_item_ptr(leaf, path->slots[0],
65 struct btrfs_free_space_header);
66 btrfs_free_space_key(leaf, header, &disk_key);
67 btrfs_disk_key_to_cpu(&location, &disk_key);
68 btrfs_release_path(path);
70 inode = btrfs_iget(root->fs_info->sb, &location, root, NULL);
72 return ERR_PTR(-ENOENT);
75 if (is_bad_inode(inode)) {
77 return ERR_PTR(-ENOENT);
80 mapping_set_gfp_mask(inode->i_mapping,
81 mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
86 struct inode *lookup_free_space_inode(struct btrfs_root *root,
87 struct btrfs_block_group_cache
88 *block_group, struct btrfs_path *path)
90 struct inode *inode = NULL;
91 u32 flags = BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
93 spin_lock(&block_group->lock);
94 if (block_group->inode)
95 inode = igrab(block_group->inode);
96 spin_unlock(&block_group->lock);
100 inode = __lookup_free_space_inode(root, path,
101 block_group->key.objectid);
105 spin_lock(&block_group->lock);
106 if (!((BTRFS_I(inode)->flags & flags) == flags)) {
107 btrfs_info(root->fs_info,
108 "Old style space inode found, converting.");
109 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM |
110 BTRFS_INODE_NODATACOW;
111 block_group->disk_cache_state = BTRFS_DC_CLEAR;
114 if (!block_group->iref) {
115 block_group->inode = igrab(inode);
116 block_group->iref = 1;
118 spin_unlock(&block_group->lock);
123 static int __create_free_space_inode(struct btrfs_root *root,
124 struct btrfs_trans_handle *trans,
125 struct btrfs_path *path,
128 struct btrfs_key key;
129 struct btrfs_disk_key disk_key;
130 struct btrfs_free_space_header *header;
131 struct btrfs_inode_item *inode_item;
132 struct extent_buffer *leaf;
133 u64 flags = BTRFS_INODE_NOCOMPRESS | BTRFS_INODE_PREALLOC;
136 ret = btrfs_insert_empty_inode(trans, root, path, ino);
140 /* We inline crc's for the free disk space cache */
141 if (ino != BTRFS_FREE_INO_OBJECTID)
142 flags |= BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
144 leaf = path->nodes[0];
145 inode_item = btrfs_item_ptr(leaf, path->slots[0],
146 struct btrfs_inode_item);
147 btrfs_item_key(leaf, &disk_key, path->slots[0]);
148 memset_extent_buffer(leaf, 0, (unsigned long)inode_item,
149 sizeof(*inode_item));
150 btrfs_set_inode_generation(leaf, inode_item, trans->transid);
151 btrfs_set_inode_size(leaf, inode_item, 0);
152 btrfs_set_inode_nbytes(leaf, inode_item, 0);
153 btrfs_set_inode_uid(leaf, inode_item, 0);
154 btrfs_set_inode_gid(leaf, inode_item, 0);
155 btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600);
156 btrfs_set_inode_flags(leaf, inode_item, flags);
157 btrfs_set_inode_nlink(leaf, inode_item, 1);
158 btrfs_set_inode_transid(leaf, inode_item, trans->transid);
159 btrfs_set_inode_block_group(leaf, inode_item, offset);
160 btrfs_mark_buffer_dirty(leaf);
161 btrfs_release_path(path);
163 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
167 ret = btrfs_insert_empty_item(trans, root, path, &key,
168 sizeof(struct btrfs_free_space_header));
170 btrfs_release_path(path);
173 leaf = path->nodes[0];
174 header = btrfs_item_ptr(leaf, path->slots[0],
175 struct btrfs_free_space_header);
176 memset_extent_buffer(leaf, 0, (unsigned long)header, sizeof(*header));
177 btrfs_set_free_space_key(leaf, header, &disk_key);
178 btrfs_mark_buffer_dirty(leaf);
179 btrfs_release_path(path);
184 int create_free_space_inode(struct btrfs_root *root,
185 struct btrfs_trans_handle *trans,
186 struct btrfs_block_group_cache *block_group,
187 struct btrfs_path *path)
192 ret = btrfs_find_free_objectid(root, &ino);
196 return __create_free_space_inode(root, trans, path, ino,
197 block_group->key.objectid);
200 int btrfs_check_trunc_cache_free_space(struct btrfs_root *root,
201 struct btrfs_block_rsv *rsv)
206 /* 1 for slack space, 1 for updating the inode */
207 needed_bytes = btrfs_calc_trunc_metadata_size(root, 1) +
208 btrfs_calc_trans_metadata_size(root, 1);
210 spin_lock(&rsv->lock);
211 if (rsv->reserved < needed_bytes)
215 spin_unlock(&rsv->lock);
219 int btrfs_truncate_free_space_cache(struct btrfs_root *root,
220 struct btrfs_trans_handle *trans,
225 btrfs_i_size_write(inode, 0);
226 truncate_pagecache(inode, 0);
229 * We don't need an orphan item because truncating the free space cache
230 * will never be split across transactions.
232 ret = btrfs_truncate_inode_items(trans, root, inode,
233 0, BTRFS_EXTENT_DATA_KEY);
235 btrfs_abort_transaction(trans, root, ret);
239 ret = btrfs_update_inode(trans, root, inode);
241 btrfs_abort_transaction(trans, root, ret);
246 static int readahead_cache(struct inode *inode)
248 struct file_ra_state *ra;
249 unsigned long last_index;
251 ra = kzalloc(sizeof(*ra), GFP_NOFS);
255 file_ra_state_init(ra, inode->i_mapping);
256 last_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT;
258 page_cache_sync_readahead(inode->i_mapping, ra, NULL, 0, last_index);
269 struct btrfs_root *root;
273 unsigned check_crcs:1;
276 static int io_ctl_init(struct io_ctl *io_ctl, struct inode *inode,
277 struct btrfs_root *root)
279 memset(io_ctl, 0, sizeof(struct io_ctl));
280 io_ctl->num_pages = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >>
282 io_ctl->pages = kzalloc(sizeof(struct page *) * io_ctl->num_pages,
287 if (btrfs_ino(inode) != BTRFS_FREE_INO_OBJECTID)
288 io_ctl->check_crcs = 1;
292 static void io_ctl_free(struct io_ctl *io_ctl)
294 kfree(io_ctl->pages);
297 static void io_ctl_unmap_page(struct io_ctl *io_ctl)
300 kunmap(io_ctl->page);
306 static void io_ctl_map_page(struct io_ctl *io_ctl, int clear)
308 ASSERT(io_ctl->index < io_ctl->num_pages);
309 io_ctl->page = io_ctl->pages[io_ctl->index++];
310 io_ctl->cur = kmap(io_ctl->page);
311 io_ctl->orig = io_ctl->cur;
312 io_ctl->size = PAGE_CACHE_SIZE;
314 memset(io_ctl->cur, 0, PAGE_CACHE_SIZE);
317 static void io_ctl_drop_pages(struct io_ctl *io_ctl)
321 io_ctl_unmap_page(io_ctl);
323 for (i = 0; i < io_ctl->num_pages; i++) {
324 if (io_ctl->pages[i]) {
325 ClearPageChecked(io_ctl->pages[i]);
326 unlock_page(io_ctl->pages[i]);
327 page_cache_release(io_ctl->pages[i]);
332 static int io_ctl_prepare_pages(struct io_ctl *io_ctl, struct inode *inode,
336 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
339 for (i = 0; i < io_ctl->num_pages; i++) {
340 page = find_or_create_page(inode->i_mapping, i, mask);
342 io_ctl_drop_pages(io_ctl);
345 io_ctl->pages[i] = page;
346 if (uptodate && !PageUptodate(page)) {
347 btrfs_readpage(NULL, page);
349 if (!PageUptodate(page)) {
350 printk(KERN_ERR "btrfs: error reading free "
352 io_ctl_drop_pages(io_ctl);
358 for (i = 0; i < io_ctl->num_pages; i++) {
359 clear_page_dirty_for_io(io_ctl->pages[i]);
360 set_page_extent_mapped(io_ctl->pages[i]);
366 static void io_ctl_set_generation(struct io_ctl *io_ctl, u64 generation)
370 io_ctl_map_page(io_ctl, 1);
373 * Skip the csum areas. If we don't check crcs then we just have a
374 * 64bit chunk at the front of the first page.
376 if (io_ctl->check_crcs) {
377 io_ctl->cur += (sizeof(u32) * io_ctl->num_pages);
378 io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages);
380 io_ctl->cur += sizeof(u64);
381 io_ctl->size -= sizeof(u64) * 2;
385 *val = cpu_to_le64(generation);
386 io_ctl->cur += sizeof(u64);
389 static int io_ctl_check_generation(struct io_ctl *io_ctl, u64 generation)
394 * Skip the crc area. If we don't check crcs then we just have a 64bit
395 * chunk at the front of the first page.
397 if (io_ctl->check_crcs) {
398 io_ctl->cur += sizeof(u32) * io_ctl->num_pages;
399 io_ctl->size -= sizeof(u64) +
400 (sizeof(u32) * io_ctl->num_pages);
402 io_ctl->cur += sizeof(u64);
403 io_ctl->size -= sizeof(u64) * 2;
407 if (le64_to_cpu(*gen) != generation) {
408 printk_ratelimited(KERN_ERR "btrfs: space cache generation "
409 "(%Lu) does not match inode (%Lu)\n", *gen,
411 io_ctl_unmap_page(io_ctl);
414 io_ctl->cur += sizeof(u64);
418 static void io_ctl_set_crc(struct io_ctl *io_ctl, int index)
424 if (!io_ctl->check_crcs) {
425 io_ctl_unmap_page(io_ctl);
430 offset = sizeof(u32) * io_ctl->num_pages;
432 crc = btrfs_csum_data(io_ctl->orig + offset, crc,
433 PAGE_CACHE_SIZE - offset);
434 btrfs_csum_final(crc, (char *)&crc);
435 io_ctl_unmap_page(io_ctl);
436 tmp = kmap(io_ctl->pages[0]);
439 kunmap(io_ctl->pages[0]);
442 static int io_ctl_check_crc(struct io_ctl *io_ctl, int index)
448 if (!io_ctl->check_crcs) {
449 io_ctl_map_page(io_ctl, 0);
454 offset = sizeof(u32) * io_ctl->num_pages;
456 tmp = kmap(io_ctl->pages[0]);
459 kunmap(io_ctl->pages[0]);
461 io_ctl_map_page(io_ctl, 0);
462 crc = btrfs_csum_data(io_ctl->orig + offset, crc,
463 PAGE_CACHE_SIZE - offset);
464 btrfs_csum_final(crc, (char *)&crc);
466 printk_ratelimited(KERN_ERR "btrfs: csum mismatch on free "
468 io_ctl_unmap_page(io_ctl);
475 static int io_ctl_add_entry(struct io_ctl *io_ctl, u64 offset, u64 bytes,
478 struct btrfs_free_space_entry *entry;
484 entry->offset = cpu_to_le64(offset);
485 entry->bytes = cpu_to_le64(bytes);
486 entry->type = (bitmap) ? BTRFS_FREE_SPACE_BITMAP :
487 BTRFS_FREE_SPACE_EXTENT;
488 io_ctl->cur += sizeof(struct btrfs_free_space_entry);
489 io_ctl->size -= sizeof(struct btrfs_free_space_entry);
491 if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
494 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
496 /* No more pages to map */
497 if (io_ctl->index >= io_ctl->num_pages)
500 /* map the next page */
501 io_ctl_map_page(io_ctl, 1);
505 static int io_ctl_add_bitmap(struct io_ctl *io_ctl, void *bitmap)
511 * If we aren't at the start of the current page, unmap this one and
512 * map the next one if there is any left.
514 if (io_ctl->cur != io_ctl->orig) {
515 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
516 if (io_ctl->index >= io_ctl->num_pages)
518 io_ctl_map_page(io_ctl, 0);
521 memcpy(io_ctl->cur, bitmap, PAGE_CACHE_SIZE);
522 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
523 if (io_ctl->index < io_ctl->num_pages)
524 io_ctl_map_page(io_ctl, 0);
528 static void io_ctl_zero_remaining_pages(struct io_ctl *io_ctl)
531 * If we're not on the boundary we know we've modified the page and we
532 * need to crc the page.
534 if (io_ctl->cur != io_ctl->orig)
535 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
537 io_ctl_unmap_page(io_ctl);
539 while (io_ctl->index < io_ctl->num_pages) {
540 io_ctl_map_page(io_ctl, 1);
541 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
545 static int io_ctl_read_entry(struct io_ctl *io_ctl,
546 struct btrfs_free_space *entry, u8 *type)
548 struct btrfs_free_space_entry *e;
552 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
558 entry->offset = le64_to_cpu(e->offset);
559 entry->bytes = le64_to_cpu(e->bytes);
561 io_ctl->cur += sizeof(struct btrfs_free_space_entry);
562 io_ctl->size -= sizeof(struct btrfs_free_space_entry);
564 if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
567 io_ctl_unmap_page(io_ctl);
572 static int io_ctl_read_bitmap(struct io_ctl *io_ctl,
573 struct btrfs_free_space *entry)
577 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
581 memcpy(entry->bitmap, io_ctl->cur, PAGE_CACHE_SIZE);
582 io_ctl_unmap_page(io_ctl);
588 * Since we attach pinned extents after the fact we can have contiguous sections
589 * of free space that are split up in entries. This poses a problem with the
590 * tree logging stuff since it could have allocated across what appears to be 2
591 * entries since we would have merged the entries when adding the pinned extents
592 * back to the free space cache. So run through the space cache that we just
593 * loaded and merge contiguous entries. This will make the log replay stuff not
594 * blow up and it will make for nicer allocator behavior.
596 static void merge_space_tree(struct btrfs_free_space_ctl *ctl)
598 struct btrfs_free_space *e, *prev = NULL;
602 spin_lock(&ctl->tree_lock);
603 for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
604 e = rb_entry(n, struct btrfs_free_space, offset_index);
607 if (e->bitmap || prev->bitmap)
609 if (prev->offset + prev->bytes == e->offset) {
610 unlink_free_space(ctl, prev);
611 unlink_free_space(ctl, e);
612 prev->bytes += e->bytes;
613 kmem_cache_free(btrfs_free_space_cachep, e);
614 link_free_space(ctl, prev);
616 spin_unlock(&ctl->tree_lock);
622 spin_unlock(&ctl->tree_lock);
625 static int __load_free_space_cache(struct btrfs_root *root, struct inode *inode,
626 struct btrfs_free_space_ctl *ctl,
627 struct btrfs_path *path, u64 offset)
629 struct btrfs_free_space_header *header;
630 struct extent_buffer *leaf;
631 struct io_ctl io_ctl;
632 struct btrfs_key key;
633 struct btrfs_free_space *e, *n;
634 struct list_head bitmaps;
641 INIT_LIST_HEAD(&bitmaps);
643 /* Nothing in the space cache, goodbye */
644 if (!i_size_read(inode))
647 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
651 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
655 btrfs_release_path(path);
661 leaf = path->nodes[0];
662 header = btrfs_item_ptr(leaf, path->slots[0],
663 struct btrfs_free_space_header);
664 num_entries = btrfs_free_space_entries(leaf, header);
665 num_bitmaps = btrfs_free_space_bitmaps(leaf, header);
666 generation = btrfs_free_space_generation(leaf, header);
667 btrfs_release_path(path);
669 if (BTRFS_I(inode)->generation != generation) {
670 btrfs_err(root->fs_info,
671 "free space inode generation (%llu) "
672 "did not match free space cache generation (%llu)",
673 BTRFS_I(inode)->generation, generation);
680 ret = io_ctl_init(&io_ctl, inode, root);
684 ret = readahead_cache(inode);
688 ret = io_ctl_prepare_pages(&io_ctl, inode, 1);
692 ret = io_ctl_check_crc(&io_ctl, 0);
696 ret = io_ctl_check_generation(&io_ctl, generation);
700 while (num_entries) {
701 e = kmem_cache_zalloc(btrfs_free_space_cachep,
706 ret = io_ctl_read_entry(&io_ctl, e, &type);
708 kmem_cache_free(btrfs_free_space_cachep, e);
713 kmem_cache_free(btrfs_free_space_cachep, e);
717 if (type == BTRFS_FREE_SPACE_EXTENT) {
718 spin_lock(&ctl->tree_lock);
719 ret = link_free_space(ctl, e);
720 spin_unlock(&ctl->tree_lock);
722 btrfs_err(root->fs_info,
723 "Duplicate entries in free space cache, dumping");
724 kmem_cache_free(btrfs_free_space_cachep, e);
730 e->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
733 btrfs_free_space_cachep, e);
736 spin_lock(&ctl->tree_lock);
737 ret = link_free_space(ctl, e);
738 ctl->total_bitmaps++;
739 ctl->op->recalc_thresholds(ctl);
740 spin_unlock(&ctl->tree_lock);
742 btrfs_err(root->fs_info,
743 "Duplicate entries in free space cache, dumping");
744 kmem_cache_free(btrfs_free_space_cachep, e);
747 list_add_tail(&e->list, &bitmaps);
753 io_ctl_unmap_page(&io_ctl);
756 * We add the bitmaps at the end of the entries in order that
757 * the bitmap entries are added to the cache.
759 list_for_each_entry_safe(e, n, &bitmaps, list) {
760 list_del_init(&e->list);
761 ret = io_ctl_read_bitmap(&io_ctl, e);
766 io_ctl_drop_pages(&io_ctl);
767 merge_space_tree(ctl);
770 io_ctl_free(&io_ctl);
773 io_ctl_drop_pages(&io_ctl);
774 __btrfs_remove_free_space_cache(ctl);
778 int load_free_space_cache(struct btrfs_fs_info *fs_info,
779 struct btrfs_block_group_cache *block_group)
781 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
782 struct btrfs_root *root = fs_info->tree_root;
784 struct btrfs_path *path;
787 u64 used = btrfs_block_group_used(&block_group->item);
790 * If this block group has been marked to be cleared for one reason or
791 * another then we can't trust the on disk cache, so just return.
793 spin_lock(&block_group->lock);
794 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
795 spin_unlock(&block_group->lock);
798 spin_unlock(&block_group->lock);
800 path = btrfs_alloc_path();
803 path->search_commit_root = 1;
804 path->skip_locking = 1;
806 inode = lookup_free_space_inode(root, block_group, path);
808 btrfs_free_path(path);
812 /* We may have converted the inode and made the cache invalid. */
813 spin_lock(&block_group->lock);
814 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
815 spin_unlock(&block_group->lock);
816 btrfs_free_path(path);
819 spin_unlock(&block_group->lock);
821 ret = __load_free_space_cache(fs_info->tree_root, inode, ctl,
822 path, block_group->key.objectid);
823 btrfs_free_path(path);
827 spin_lock(&ctl->tree_lock);
828 matched = (ctl->free_space == (block_group->key.offset - used -
829 block_group->bytes_super));
830 spin_unlock(&ctl->tree_lock);
833 __btrfs_remove_free_space_cache(ctl);
834 btrfs_err(fs_info, "block group %llu has wrong amount of free space",
835 block_group->key.objectid);
840 /* This cache is bogus, make sure it gets cleared */
841 spin_lock(&block_group->lock);
842 block_group->disk_cache_state = BTRFS_DC_CLEAR;
843 spin_unlock(&block_group->lock);
846 btrfs_err(fs_info, "failed to load free space cache for block group %llu",
847 block_group->key.objectid);
855 * __btrfs_write_out_cache - write out cached info to an inode
856 * @root - the root the inode belongs to
857 * @ctl - the free space cache we are going to write out
858 * @block_group - the block_group for this cache if it belongs to a block_group
859 * @trans - the trans handle
860 * @path - the path to use
861 * @offset - the offset for the key we'll insert
863 * This function writes out a free space cache struct to disk for quick recovery
864 * on mount. This will return 0 if it was successfull in writing the cache out,
865 * and -1 if it was not.
867 static int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
868 struct btrfs_free_space_ctl *ctl,
869 struct btrfs_block_group_cache *block_group,
870 struct btrfs_trans_handle *trans,
871 struct btrfs_path *path, u64 offset)
873 struct btrfs_free_space_header *header;
874 struct extent_buffer *leaf;
875 struct rb_node *node;
876 struct list_head *pos, *n;
877 struct extent_state *cached_state = NULL;
878 struct btrfs_free_cluster *cluster = NULL;
879 struct extent_io_tree *unpin = NULL;
880 struct io_ctl io_ctl;
881 struct list_head bitmap_list;
882 struct btrfs_key key;
883 u64 start, extent_start, extent_end, len;
889 INIT_LIST_HEAD(&bitmap_list);
891 if (!i_size_read(inode))
894 ret = io_ctl_init(&io_ctl, inode, root);
898 /* Get the cluster for this block_group if it exists */
899 if (block_group && !list_empty(&block_group->cluster_list))
900 cluster = list_entry(block_group->cluster_list.next,
901 struct btrfs_free_cluster,
904 /* Lock all pages first so we can lock the extent safely. */
905 io_ctl_prepare_pages(&io_ctl, inode, 0);
907 lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
910 node = rb_first(&ctl->free_space_offset);
911 if (!node && cluster) {
912 node = rb_first(&cluster->root);
916 /* Make sure we can fit our crcs into the first page */
917 if (io_ctl.check_crcs &&
918 (io_ctl.num_pages * sizeof(u32)) >= PAGE_CACHE_SIZE)
921 io_ctl_set_generation(&io_ctl, trans->transid);
923 /* Write out the extent entries */
925 struct btrfs_free_space *e;
927 e = rb_entry(node, struct btrfs_free_space, offset_index);
930 ret = io_ctl_add_entry(&io_ctl, e->offset, e->bytes,
936 list_add_tail(&e->list, &bitmap_list);
939 node = rb_next(node);
940 if (!node && cluster) {
941 node = rb_first(&cluster->root);
947 * We want to add any pinned extents to our free space cache
948 * so we don't leak the space
952 * We shouldn't have switched the pinned extents yet so this is the
955 unpin = root->fs_info->pinned_extents;
958 start = block_group->key.objectid;
960 while (block_group && (start < block_group->key.objectid +
961 block_group->key.offset)) {
962 ret = find_first_extent_bit(unpin, start,
963 &extent_start, &extent_end,
970 /* This pinned extent is out of our range */
971 if (extent_start >= block_group->key.objectid +
972 block_group->key.offset)
975 extent_start = max(extent_start, start);
976 extent_end = min(block_group->key.objectid +
977 block_group->key.offset, extent_end + 1);
978 len = extent_end - extent_start;
981 ret = io_ctl_add_entry(&io_ctl, extent_start, len, NULL);
988 /* Write out the bitmaps */
989 list_for_each_safe(pos, n, &bitmap_list) {
990 struct btrfs_free_space *entry =
991 list_entry(pos, struct btrfs_free_space, list);
993 ret = io_ctl_add_bitmap(&io_ctl, entry->bitmap);
996 list_del_init(&entry->list);
999 /* Zero out the rest of the pages just to make sure */
1000 io_ctl_zero_remaining_pages(&io_ctl);
1002 ret = btrfs_dirty_pages(root, inode, io_ctl.pages, io_ctl.num_pages,
1003 0, i_size_read(inode), &cached_state);
1004 io_ctl_drop_pages(&io_ctl);
1005 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1006 i_size_read(inode) - 1, &cached_state, GFP_NOFS);
1011 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
1013 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
1014 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL,
1019 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
1020 key.offset = offset;
1023 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1025 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
1026 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL,
1030 leaf = path->nodes[0];
1032 struct btrfs_key found_key;
1033 ASSERT(path->slots[0]);
1035 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1036 if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID ||
1037 found_key.offset != offset) {
1038 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0,
1040 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0,
1042 btrfs_release_path(path);
1047 BTRFS_I(inode)->generation = trans->transid;
1048 header = btrfs_item_ptr(leaf, path->slots[0],
1049 struct btrfs_free_space_header);
1050 btrfs_set_free_space_entries(leaf, header, entries);
1051 btrfs_set_free_space_bitmaps(leaf, header, bitmaps);
1052 btrfs_set_free_space_generation(leaf, header, trans->transid);
1053 btrfs_mark_buffer_dirty(leaf);
1054 btrfs_release_path(path);
1058 io_ctl_free(&io_ctl);
1060 invalidate_inode_pages2(inode->i_mapping);
1061 BTRFS_I(inode)->generation = 0;
1063 btrfs_update_inode(trans, root, inode);
1067 list_for_each_safe(pos, n, &bitmap_list) {
1068 struct btrfs_free_space *entry =
1069 list_entry(pos, struct btrfs_free_space, list);
1070 list_del_init(&entry->list);
1072 io_ctl_drop_pages(&io_ctl);
1073 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1074 i_size_read(inode) - 1, &cached_state, GFP_NOFS);
1078 int btrfs_write_out_cache(struct btrfs_root *root,
1079 struct btrfs_trans_handle *trans,
1080 struct btrfs_block_group_cache *block_group,
1081 struct btrfs_path *path)
1083 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1084 struct inode *inode;
1087 root = root->fs_info->tree_root;
1089 spin_lock(&block_group->lock);
1090 if (block_group->disk_cache_state < BTRFS_DC_SETUP) {
1091 spin_unlock(&block_group->lock);
1094 spin_unlock(&block_group->lock);
1096 inode = lookup_free_space_inode(root, block_group, path);
1100 ret = __btrfs_write_out_cache(root, inode, ctl, block_group, trans,
1101 path, block_group->key.objectid);
1103 spin_lock(&block_group->lock);
1104 block_group->disk_cache_state = BTRFS_DC_ERROR;
1105 spin_unlock(&block_group->lock);
1108 btrfs_err(root->fs_info,
1109 "failed to write free space cache for block group %llu",
1110 block_group->key.objectid);
1118 static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit,
1121 ASSERT(offset >= bitmap_start);
1122 offset -= bitmap_start;
1123 return (unsigned long)(div_u64(offset, unit));
1126 static inline unsigned long bytes_to_bits(u64 bytes, u32 unit)
1128 return (unsigned long)(div_u64(bytes, unit));
1131 static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl,
1135 u64 bytes_per_bitmap;
1137 bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit;
1138 bitmap_start = offset - ctl->start;
1139 bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap);
1140 bitmap_start *= bytes_per_bitmap;
1141 bitmap_start += ctl->start;
1143 return bitmap_start;
1146 static int tree_insert_offset(struct rb_root *root, u64 offset,
1147 struct rb_node *node, int bitmap)
1149 struct rb_node **p = &root->rb_node;
1150 struct rb_node *parent = NULL;
1151 struct btrfs_free_space *info;
1155 info = rb_entry(parent, struct btrfs_free_space, offset_index);
1157 if (offset < info->offset) {
1159 } else if (offset > info->offset) {
1160 p = &(*p)->rb_right;
1163 * we could have a bitmap entry and an extent entry
1164 * share the same offset. If this is the case, we want
1165 * the extent entry to always be found first if we do a
1166 * linear search through the tree, since we want to have
1167 * the quickest allocation time, and allocating from an
1168 * extent is faster than allocating from a bitmap. So
1169 * if we're inserting a bitmap and we find an entry at
1170 * this offset, we want to go right, or after this entry
1171 * logically. If we are inserting an extent and we've
1172 * found a bitmap, we want to go left, or before
1180 p = &(*p)->rb_right;
1182 if (!info->bitmap) {
1191 rb_link_node(node, parent, p);
1192 rb_insert_color(node, root);
1198 * searches the tree for the given offset.
1200 * fuzzy - If this is set, then we are trying to make an allocation, and we just
1201 * want a section that has at least bytes size and comes at or after the given
1204 static struct btrfs_free_space *
1205 tree_search_offset(struct btrfs_free_space_ctl *ctl,
1206 u64 offset, int bitmap_only, int fuzzy)
1208 struct rb_node *n = ctl->free_space_offset.rb_node;
1209 struct btrfs_free_space *entry, *prev = NULL;
1211 /* find entry that is closest to the 'offset' */
1218 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1221 if (offset < entry->offset)
1223 else if (offset > entry->offset)
1236 * bitmap entry and extent entry may share same offset,
1237 * in that case, bitmap entry comes after extent entry.
1242 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1243 if (entry->offset != offset)
1246 WARN_ON(!entry->bitmap);
1249 if (entry->bitmap) {
1251 * if previous extent entry covers the offset,
1252 * we should return it instead of the bitmap entry
1254 n = rb_prev(&entry->offset_index);
1256 prev = rb_entry(n, struct btrfs_free_space,
1258 if (!prev->bitmap &&
1259 prev->offset + prev->bytes > offset)
1269 /* find last entry before the 'offset' */
1271 if (entry->offset > offset) {
1272 n = rb_prev(&entry->offset_index);
1274 entry = rb_entry(n, struct btrfs_free_space,
1276 ASSERT(entry->offset <= offset);
1285 if (entry->bitmap) {
1286 n = rb_prev(&entry->offset_index);
1288 prev = rb_entry(n, struct btrfs_free_space,
1290 if (!prev->bitmap &&
1291 prev->offset + prev->bytes > offset)
1294 if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset)
1296 } else if (entry->offset + entry->bytes > offset)
1303 if (entry->bitmap) {
1304 if (entry->offset + BITS_PER_BITMAP *
1308 if (entry->offset + entry->bytes > offset)
1312 n = rb_next(&entry->offset_index);
1315 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1321 __unlink_free_space(struct btrfs_free_space_ctl *ctl,
1322 struct btrfs_free_space *info)
1324 rb_erase(&info->offset_index, &ctl->free_space_offset);
1325 ctl->free_extents--;
1328 static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
1329 struct btrfs_free_space *info)
1331 __unlink_free_space(ctl, info);
1332 ctl->free_space -= info->bytes;
1335 static int link_free_space(struct btrfs_free_space_ctl *ctl,
1336 struct btrfs_free_space *info)
1340 ASSERT(info->bytes || info->bitmap);
1341 ret = tree_insert_offset(&ctl->free_space_offset, info->offset,
1342 &info->offset_index, (info->bitmap != NULL));
1346 ctl->free_space += info->bytes;
1347 ctl->free_extents++;
1351 static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
1353 struct btrfs_block_group_cache *block_group = ctl->private;
1357 u64 size = block_group->key.offset;
1358 u64 bytes_per_bg = BITS_PER_BITMAP * ctl->unit;
1359 int max_bitmaps = div64_u64(size + bytes_per_bg - 1, bytes_per_bg);
1361 max_bitmaps = max(max_bitmaps, 1);
1363 ASSERT(ctl->total_bitmaps <= max_bitmaps);
1366 * The goal is to keep the total amount of memory used per 1gb of space
1367 * at or below 32k, so we need to adjust how much memory we allow to be
1368 * used by extent based free space tracking
1370 if (size < 1024 * 1024 * 1024)
1371 max_bytes = MAX_CACHE_BYTES_PER_GIG;
1373 max_bytes = MAX_CACHE_BYTES_PER_GIG *
1374 div64_u64(size, 1024 * 1024 * 1024);
1377 * we want to account for 1 more bitmap than what we have so we can make
1378 * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as
1379 * we add more bitmaps.
1381 bitmap_bytes = (ctl->total_bitmaps + 1) * PAGE_CACHE_SIZE;
1383 if (bitmap_bytes >= max_bytes) {
1384 ctl->extents_thresh = 0;
1389 * we want the extent entry threshold to always be at most 1/2 the maxw
1390 * bytes we can have, or whatever is less than that.
1392 extent_bytes = max_bytes - bitmap_bytes;
1393 extent_bytes = min_t(u64, extent_bytes, div64_u64(max_bytes, 2));
1395 ctl->extents_thresh =
1396 div64_u64(extent_bytes, (sizeof(struct btrfs_free_space)));
1399 static inline void __bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1400 struct btrfs_free_space *info,
1401 u64 offset, u64 bytes)
1403 unsigned long start, count;
1405 start = offset_to_bit(info->offset, ctl->unit, offset);
1406 count = bytes_to_bits(bytes, ctl->unit);
1407 ASSERT(start + count <= BITS_PER_BITMAP);
1409 bitmap_clear(info->bitmap, start, count);
1411 info->bytes -= bytes;
1414 static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1415 struct btrfs_free_space *info, u64 offset,
1418 __bitmap_clear_bits(ctl, info, offset, bytes);
1419 ctl->free_space -= bytes;
1422 static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl,
1423 struct btrfs_free_space *info, u64 offset,
1426 unsigned long start, count;
1428 start = offset_to_bit(info->offset, ctl->unit, offset);
1429 count = bytes_to_bits(bytes, ctl->unit);
1430 ASSERT(start + count <= BITS_PER_BITMAP);
1432 bitmap_set(info->bitmap, start, count);
1434 info->bytes += bytes;
1435 ctl->free_space += bytes;
1439 * If we can not find suitable extent, we will use bytes to record
1440 * the size of the max extent.
1442 static int search_bitmap(struct btrfs_free_space_ctl *ctl,
1443 struct btrfs_free_space *bitmap_info, u64 *offset,
1446 unsigned long found_bits = 0;
1447 unsigned long max_bits = 0;
1448 unsigned long bits, i;
1449 unsigned long next_zero;
1450 unsigned long extent_bits;
1452 i = offset_to_bit(bitmap_info->offset, ctl->unit,
1453 max_t(u64, *offset, bitmap_info->offset));
1454 bits = bytes_to_bits(*bytes, ctl->unit);
1456 for_each_set_bit_from(i, bitmap_info->bitmap, BITS_PER_BITMAP) {
1457 next_zero = find_next_zero_bit(bitmap_info->bitmap,
1458 BITS_PER_BITMAP, i);
1459 extent_bits = next_zero - i;
1460 if (extent_bits >= bits) {
1461 found_bits = extent_bits;
1463 } else if (extent_bits > max_bits) {
1464 max_bits = extent_bits;
1470 *offset = (u64)(i * ctl->unit) + bitmap_info->offset;
1471 *bytes = (u64)(found_bits) * ctl->unit;
1475 *bytes = (u64)(max_bits) * ctl->unit;
1479 /* Cache the size of the max extent in bytes */
1480 static struct btrfs_free_space *
1481 find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes,
1482 unsigned long align, u64 *max_extent_size)
1484 struct btrfs_free_space *entry;
1485 struct rb_node *node;
1490 if (!ctl->free_space_offset.rb_node)
1493 entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset), 0, 1);
1497 for (node = &entry->offset_index; node; node = rb_next(node)) {
1498 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1499 if (entry->bytes < *bytes) {
1500 if (entry->bytes > *max_extent_size)
1501 *max_extent_size = entry->bytes;
1505 /* make sure the space returned is big enough
1506 * to match our requested alignment
1508 if (*bytes >= align) {
1509 tmp = entry->offset - ctl->start + align - 1;
1511 tmp = tmp * align + ctl->start;
1512 align_off = tmp - entry->offset;
1515 tmp = entry->offset;
1518 if (entry->bytes < *bytes + align_off) {
1519 if (entry->bytes > *max_extent_size)
1520 *max_extent_size = entry->bytes;
1524 if (entry->bitmap) {
1527 ret = search_bitmap(ctl, entry, &tmp, &size);
1532 } else if (size > *max_extent_size) {
1533 *max_extent_size = size;
1539 *bytes = entry->bytes - align_off;
1546 static void add_new_bitmap(struct btrfs_free_space_ctl *ctl,
1547 struct btrfs_free_space *info, u64 offset)
1549 info->offset = offset_to_bitmap(ctl, offset);
1551 INIT_LIST_HEAD(&info->list);
1552 link_free_space(ctl, info);
1553 ctl->total_bitmaps++;
1555 ctl->op->recalc_thresholds(ctl);
1558 static void free_bitmap(struct btrfs_free_space_ctl *ctl,
1559 struct btrfs_free_space *bitmap_info)
1561 unlink_free_space(ctl, bitmap_info);
1562 kfree(bitmap_info->bitmap);
1563 kmem_cache_free(btrfs_free_space_cachep, bitmap_info);
1564 ctl->total_bitmaps--;
1565 ctl->op->recalc_thresholds(ctl);
1568 static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl,
1569 struct btrfs_free_space *bitmap_info,
1570 u64 *offset, u64 *bytes)
1573 u64 search_start, search_bytes;
1577 end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1;
1580 * We need to search for bits in this bitmap. We could only cover some
1581 * of the extent in this bitmap thanks to how we add space, so we need
1582 * to search for as much as it as we can and clear that amount, and then
1583 * go searching for the next bit.
1585 search_start = *offset;
1586 search_bytes = ctl->unit;
1587 search_bytes = min(search_bytes, end - search_start + 1);
1588 ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes);
1589 if (ret < 0 || search_start != *offset)
1592 /* We may have found more bits than what we need */
1593 search_bytes = min(search_bytes, *bytes);
1595 /* Cannot clear past the end of the bitmap */
1596 search_bytes = min(search_bytes, end - search_start + 1);
1598 bitmap_clear_bits(ctl, bitmap_info, search_start, search_bytes);
1599 *offset += search_bytes;
1600 *bytes -= search_bytes;
1603 struct rb_node *next = rb_next(&bitmap_info->offset_index);
1604 if (!bitmap_info->bytes)
1605 free_bitmap(ctl, bitmap_info);
1608 * no entry after this bitmap, but we still have bytes to
1609 * remove, so something has gone wrong.
1614 bitmap_info = rb_entry(next, struct btrfs_free_space,
1618 * if the next entry isn't a bitmap we need to return to let the
1619 * extent stuff do its work.
1621 if (!bitmap_info->bitmap)
1625 * Ok the next item is a bitmap, but it may not actually hold
1626 * the information for the rest of this free space stuff, so
1627 * look for it, and if we don't find it return so we can try
1628 * everything over again.
1630 search_start = *offset;
1631 search_bytes = ctl->unit;
1632 ret = search_bitmap(ctl, bitmap_info, &search_start,
1634 if (ret < 0 || search_start != *offset)
1638 } else if (!bitmap_info->bytes)
1639 free_bitmap(ctl, bitmap_info);
1644 static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl,
1645 struct btrfs_free_space *info, u64 offset,
1648 u64 bytes_to_set = 0;
1651 end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit);
1653 bytes_to_set = min(end - offset, bytes);
1655 bitmap_set_bits(ctl, info, offset, bytes_to_set);
1657 return bytes_to_set;
1661 static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
1662 struct btrfs_free_space *info)
1664 struct btrfs_block_group_cache *block_group = ctl->private;
1667 * If we are below the extents threshold then we can add this as an
1668 * extent, and don't have to deal with the bitmap
1670 if (ctl->free_extents < ctl->extents_thresh) {
1672 * If this block group has some small extents we don't want to
1673 * use up all of our free slots in the cache with them, we want
1674 * to reserve them to larger extents, however if we have plent
1675 * of cache left then go ahead an dadd them, no sense in adding
1676 * the overhead of a bitmap if we don't have to.
1678 if (info->bytes <= block_group->sectorsize * 4) {
1679 if (ctl->free_extents * 2 <= ctl->extents_thresh)
1687 * The original block groups from mkfs can be really small, like 8
1688 * megabytes, so don't bother with a bitmap for those entries. However
1689 * some block groups can be smaller than what a bitmap would cover but
1690 * are still large enough that they could overflow the 32k memory limit,
1691 * so allow those block groups to still be allowed to have a bitmap
1694 if (((BITS_PER_BITMAP * ctl->unit) >> 1) > block_group->key.offset)
1700 static struct btrfs_free_space_op free_space_op = {
1701 .recalc_thresholds = recalculate_thresholds,
1702 .use_bitmap = use_bitmap,
1705 static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl,
1706 struct btrfs_free_space *info)
1708 struct btrfs_free_space *bitmap_info;
1709 struct btrfs_block_group_cache *block_group = NULL;
1711 u64 bytes, offset, bytes_added;
1714 bytes = info->bytes;
1715 offset = info->offset;
1717 if (!ctl->op->use_bitmap(ctl, info))
1720 if (ctl->op == &free_space_op)
1721 block_group = ctl->private;
1724 * Since we link bitmaps right into the cluster we need to see if we
1725 * have a cluster here, and if so and it has our bitmap we need to add
1726 * the free space to that bitmap.
1728 if (block_group && !list_empty(&block_group->cluster_list)) {
1729 struct btrfs_free_cluster *cluster;
1730 struct rb_node *node;
1731 struct btrfs_free_space *entry;
1733 cluster = list_entry(block_group->cluster_list.next,
1734 struct btrfs_free_cluster,
1736 spin_lock(&cluster->lock);
1737 node = rb_first(&cluster->root);
1739 spin_unlock(&cluster->lock);
1740 goto no_cluster_bitmap;
1743 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1744 if (!entry->bitmap) {
1745 spin_unlock(&cluster->lock);
1746 goto no_cluster_bitmap;
1749 if (entry->offset == offset_to_bitmap(ctl, offset)) {
1750 bytes_added = add_bytes_to_bitmap(ctl, entry,
1752 bytes -= bytes_added;
1753 offset += bytes_added;
1755 spin_unlock(&cluster->lock);
1763 bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
1770 bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes);
1771 bytes -= bytes_added;
1772 offset += bytes_added;
1782 if (info && info->bitmap) {
1783 add_new_bitmap(ctl, info, offset);
1788 spin_unlock(&ctl->tree_lock);
1790 /* no pre-allocated info, allocate a new one */
1792 info = kmem_cache_zalloc(btrfs_free_space_cachep,
1795 spin_lock(&ctl->tree_lock);
1801 /* allocate the bitmap */
1802 info->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
1803 spin_lock(&ctl->tree_lock);
1804 if (!info->bitmap) {
1814 kfree(info->bitmap);
1815 kmem_cache_free(btrfs_free_space_cachep, info);
1821 static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl,
1822 struct btrfs_free_space *info, bool update_stat)
1824 struct btrfs_free_space *left_info;
1825 struct btrfs_free_space *right_info;
1826 bool merged = false;
1827 u64 offset = info->offset;
1828 u64 bytes = info->bytes;
1831 * first we want to see if there is free space adjacent to the range we
1832 * are adding, if there is remove that struct and add a new one to
1833 * cover the entire range
1835 right_info = tree_search_offset(ctl, offset + bytes, 0, 0);
1836 if (right_info && rb_prev(&right_info->offset_index))
1837 left_info = rb_entry(rb_prev(&right_info->offset_index),
1838 struct btrfs_free_space, offset_index);
1840 left_info = tree_search_offset(ctl, offset - 1, 0, 0);
1842 if (right_info && !right_info->bitmap) {
1844 unlink_free_space(ctl, right_info);
1846 __unlink_free_space(ctl, right_info);
1847 info->bytes += right_info->bytes;
1848 kmem_cache_free(btrfs_free_space_cachep, right_info);
1852 if (left_info && !left_info->bitmap &&
1853 left_info->offset + left_info->bytes == offset) {
1855 unlink_free_space(ctl, left_info);
1857 __unlink_free_space(ctl, left_info);
1858 info->offset = left_info->offset;
1859 info->bytes += left_info->bytes;
1860 kmem_cache_free(btrfs_free_space_cachep, left_info);
1867 int __btrfs_add_free_space(struct btrfs_free_space_ctl *ctl,
1868 u64 offset, u64 bytes)
1870 struct btrfs_free_space *info;
1873 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
1877 info->offset = offset;
1878 info->bytes = bytes;
1880 spin_lock(&ctl->tree_lock);
1882 if (try_merge_free_space(ctl, info, true))
1886 * There was no extent directly to the left or right of this new
1887 * extent then we know we're going to have to allocate a new extent, so
1888 * before we do that see if we need to drop this into a bitmap
1890 ret = insert_into_bitmap(ctl, info);
1898 ret = link_free_space(ctl, info);
1900 kmem_cache_free(btrfs_free_space_cachep, info);
1902 spin_unlock(&ctl->tree_lock);
1905 printk(KERN_CRIT "btrfs: unable to add free space :%d\n", ret);
1906 ASSERT(ret != -EEXIST);
1912 int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group,
1913 u64 offset, u64 bytes)
1915 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1916 struct btrfs_free_space *info;
1918 bool re_search = false;
1920 spin_lock(&ctl->tree_lock);
1927 info = tree_search_offset(ctl, offset, 0, 0);
1930 * oops didn't find an extent that matched the space we wanted
1931 * to remove, look for a bitmap instead
1933 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
1937 * If we found a partial bit of our free space in a
1938 * bitmap but then couldn't find the other part this may
1939 * be a problem, so WARN about it.
1947 if (!info->bitmap) {
1948 unlink_free_space(ctl, info);
1949 if (offset == info->offset) {
1950 u64 to_free = min(bytes, info->bytes);
1952 info->bytes -= to_free;
1953 info->offset += to_free;
1955 ret = link_free_space(ctl, info);
1958 kmem_cache_free(btrfs_free_space_cachep, info);
1965 u64 old_end = info->bytes + info->offset;
1967 info->bytes = offset - info->offset;
1968 ret = link_free_space(ctl, info);
1973 /* Not enough bytes in this entry to satisfy us */
1974 if (old_end < offset + bytes) {
1975 bytes -= old_end - offset;
1978 } else if (old_end == offset + bytes) {
1982 spin_unlock(&ctl->tree_lock);
1984 ret = btrfs_add_free_space(block_group, offset + bytes,
1985 old_end - (offset + bytes));
1991 ret = remove_from_bitmap(ctl, info, &offset, &bytes);
1992 if (ret == -EAGAIN) {
1997 spin_unlock(&ctl->tree_lock);
2002 void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group,
2005 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2006 struct btrfs_free_space *info;
2010 for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
2011 info = rb_entry(n, struct btrfs_free_space, offset_index);
2012 if (info->bytes >= bytes && !block_group->ro)
2014 printk(KERN_CRIT "entry offset %llu, bytes %llu, bitmap %s\n",
2015 info->offset, info->bytes,
2016 (info->bitmap) ? "yes" : "no");
2018 printk(KERN_INFO "block group has cluster?: %s\n",
2019 list_empty(&block_group->cluster_list) ? "no" : "yes");
2020 printk(KERN_INFO "%d blocks of free space at or bigger than bytes is"
2024 void btrfs_init_free_space_ctl(struct btrfs_block_group_cache *block_group)
2026 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2028 spin_lock_init(&ctl->tree_lock);
2029 ctl->unit = block_group->sectorsize;
2030 ctl->start = block_group->key.objectid;
2031 ctl->private = block_group;
2032 ctl->op = &free_space_op;
2035 * we only want to have 32k of ram per block group for keeping
2036 * track of free space, and if we pass 1/2 of that we want to
2037 * start converting things over to using bitmaps
2039 ctl->extents_thresh = ((1024 * 32) / 2) /
2040 sizeof(struct btrfs_free_space);
2044 * for a given cluster, put all of its extents back into the free
2045 * space cache. If the block group passed doesn't match the block group
2046 * pointed to by the cluster, someone else raced in and freed the
2047 * cluster already. In that case, we just return without changing anything
2050 __btrfs_return_cluster_to_free_space(
2051 struct btrfs_block_group_cache *block_group,
2052 struct btrfs_free_cluster *cluster)
2054 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2055 struct btrfs_free_space *entry;
2056 struct rb_node *node;
2058 spin_lock(&cluster->lock);
2059 if (cluster->block_group != block_group)
2062 cluster->block_group = NULL;
2063 cluster->window_start = 0;
2064 list_del_init(&cluster->block_group_list);
2066 node = rb_first(&cluster->root);
2070 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2071 node = rb_next(&entry->offset_index);
2072 rb_erase(&entry->offset_index, &cluster->root);
2074 bitmap = (entry->bitmap != NULL);
2076 try_merge_free_space(ctl, entry, false);
2077 tree_insert_offset(&ctl->free_space_offset,
2078 entry->offset, &entry->offset_index, bitmap);
2080 cluster->root = RB_ROOT;
2083 spin_unlock(&cluster->lock);
2084 btrfs_put_block_group(block_group);
2088 static void __btrfs_remove_free_space_cache_locked(
2089 struct btrfs_free_space_ctl *ctl)
2091 struct btrfs_free_space *info;
2092 struct rb_node *node;
2094 while ((node = rb_last(&ctl->free_space_offset)) != NULL) {
2095 info = rb_entry(node, struct btrfs_free_space, offset_index);
2096 if (!info->bitmap) {
2097 unlink_free_space(ctl, info);
2098 kmem_cache_free(btrfs_free_space_cachep, info);
2100 free_bitmap(ctl, info);
2102 if (need_resched()) {
2103 spin_unlock(&ctl->tree_lock);
2105 spin_lock(&ctl->tree_lock);
2110 void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl)
2112 spin_lock(&ctl->tree_lock);
2113 __btrfs_remove_free_space_cache_locked(ctl);
2114 spin_unlock(&ctl->tree_lock);
2117 void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group)
2119 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2120 struct btrfs_free_cluster *cluster;
2121 struct list_head *head;
2123 spin_lock(&ctl->tree_lock);
2124 while ((head = block_group->cluster_list.next) !=
2125 &block_group->cluster_list) {
2126 cluster = list_entry(head, struct btrfs_free_cluster,
2129 WARN_ON(cluster->block_group != block_group);
2130 __btrfs_return_cluster_to_free_space(block_group, cluster);
2131 if (need_resched()) {
2132 spin_unlock(&ctl->tree_lock);
2134 spin_lock(&ctl->tree_lock);
2137 __btrfs_remove_free_space_cache_locked(ctl);
2138 spin_unlock(&ctl->tree_lock);
2142 u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group,
2143 u64 offset, u64 bytes, u64 empty_size,
2144 u64 *max_extent_size)
2146 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2147 struct btrfs_free_space *entry = NULL;
2148 u64 bytes_search = bytes + empty_size;
2151 u64 align_gap_len = 0;
2153 spin_lock(&ctl->tree_lock);
2154 entry = find_free_space(ctl, &offset, &bytes_search,
2155 block_group->full_stripe_len, max_extent_size);
2160 if (entry->bitmap) {
2161 bitmap_clear_bits(ctl, entry, offset, bytes);
2163 free_bitmap(ctl, entry);
2165 unlink_free_space(ctl, entry);
2166 align_gap_len = offset - entry->offset;
2167 align_gap = entry->offset;
2169 entry->offset = offset + bytes;
2170 WARN_ON(entry->bytes < bytes + align_gap_len);
2172 entry->bytes -= bytes + align_gap_len;
2174 kmem_cache_free(btrfs_free_space_cachep, entry);
2176 link_free_space(ctl, entry);
2179 spin_unlock(&ctl->tree_lock);
2182 __btrfs_add_free_space(ctl, align_gap, align_gap_len);
2187 * given a cluster, put all of its extents back into the free space
2188 * cache. If a block group is passed, this function will only free
2189 * a cluster that belongs to the passed block group.
2191 * Otherwise, it'll get a reference on the block group pointed to by the
2192 * cluster and remove the cluster from it.
2194 int btrfs_return_cluster_to_free_space(
2195 struct btrfs_block_group_cache *block_group,
2196 struct btrfs_free_cluster *cluster)
2198 struct btrfs_free_space_ctl *ctl;
2201 /* first, get a safe pointer to the block group */
2202 spin_lock(&cluster->lock);
2204 block_group = cluster->block_group;
2206 spin_unlock(&cluster->lock);
2209 } else if (cluster->block_group != block_group) {
2210 /* someone else has already freed it don't redo their work */
2211 spin_unlock(&cluster->lock);
2214 atomic_inc(&block_group->count);
2215 spin_unlock(&cluster->lock);
2217 ctl = block_group->free_space_ctl;
2219 /* now return any extents the cluster had on it */
2220 spin_lock(&ctl->tree_lock);
2221 ret = __btrfs_return_cluster_to_free_space(block_group, cluster);
2222 spin_unlock(&ctl->tree_lock);
2224 /* finally drop our ref */
2225 btrfs_put_block_group(block_group);
2229 static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group,
2230 struct btrfs_free_cluster *cluster,
2231 struct btrfs_free_space *entry,
2232 u64 bytes, u64 min_start,
2233 u64 *max_extent_size)
2235 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2237 u64 search_start = cluster->window_start;
2238 u64 search_bytes = bytes;
2241 search_start = min_start;
2242 search_bytes = bytes;
2244 err = search_bitmap(ctl, entry, &search_start, &search_bytes);
2246 if (search_bytes > *max_extent_size)
2247 *max_extent_size = search_bytes;
2252 __bitmap_clear_bits(ctl, entry, ret, bytes);
2258 * given a cluster, try to allocate 'bytes' from it, returns 0
2259 * if it couldn't find anything suitably large, or a logical disk offset
2260 * if things worked out
2262 u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group,
2263 struct btrfs_free_cluster *cluster, u64 bytes,
2264 u64 min_start, u64 *max_extent_size)
2266 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2267 struct btrfs_free_space *entry = NULL;
2268 struct rb_node *node;
2271 spin_lock(&cluster->lock);
2272 if (bytes > cluster->max_size)
2275 if (cluster->block_group != block_group)
2278 node = rb_first(&cluster->root);
2282 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2284 if (entry->bytes < bytes && entry->bytes > *max_extent_size)
2285 *max_extent_size = entry->bytes;
2287 if (entry->bytes < bytes ||
2288 (!entry->bitmap && entry->offset < min_start)) {
2289 node = rb_next(&entry->offset_index);
2292 entry = rb_entry(node, struct btrfs_free_space,
2297 if (entry->bitmap) {
2298 ret = btrfs_alloc_from_bitmap(block_group,
2299 cluster, entry, bytes,
2300 cluster->window_start,
2303 node = rb_next(&entry->offset_index);
2306 entry = rb_entry(node, struct btrfs_free_space,
2310 cluster->window_start += bytes;
2312 ret = entry->offset;
2314 entry->offset += bytes;
2315 entry->bytes -= bytes;
2318 if (entry->bytes == 0)
2319 rb_erase(&entry->offset_index, &cluster->root);
2323 spin_unlock(&cluster->lock);
2328 spin_lock(&ctl->tree_lock);
2330 ctl->free_space -= bytes;
2331 if (entry->bytes == 0) {
2332 ctl->free_extents--;
2333 if (entry->bitmap) {
2334 kfree(entry->bitmap);
2335 ctl->total_bitmaps--;
2336 ctl->op->recalc_thresholds(ctl);
2338 kmem_cache_free(btrfs_free_space_cachep, entry);
2341 spin_unlock(&ctl->tree_lock);
2346 static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group,
2347 struct btrfs_free_space *entry,
2348 struct btrfs_free_cluster *cluster,
2349 u64 offset, u64 bytes,
2350 u64 cont1_bytes, u64 min_bytes)
2352 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2353 unsigned long next_zero;
2355 unsigned long want_bits;
2356 unsigned long min_bits;
2357 unsigned long found_bits;
2358 unsigned long start = 0;
2359 unsigned long total_found = 0;
2362 i = offset_to_bit(entry->offset, ctl->unit,
2363 max_t(u64, offset, entry->offset));
2364 want_bits = bytes_to_bits(bytes, ctl->unit);
2365 min_bits = bytes_to_bits(min_bytes, ctl->unit);
2369 for_each_set_bit_from(i, entry->bitmap, BITS_PER_BITMAP) {
2370 next_zero = find_next_zero_bit(entry->bitmap,
2371 BITS_PER_BITMAP, i);
2372 if (next_zero - i >= min_bits) {
2373 found_bits = next_zero - i;
2384 cluster->max_size = 0;
2387 total_found += found_bits;
2389 if (cluster->max_size < found_bits * ctl->unit)
2390 cluster->max_size = found_bits * ctl->unit;
2392 if (total_found < want_bits || cluster->max_size < cont1_bytes) {
2397 cluster->window_start = start * ctl->unit + entry->offset;
2398 rb_erase(&entry->offset_index, &ctl->free_space_offset);
2399 ret = tree_insert_offset(&cluster->root, entry->offset,
2400 &entry->offset_index, 1);
2401 ASSERT(!ret); /* -EEXIST; Logic error */
2403 trace_btrfs_setup_cluster(block_group, cluster,
2404 total_found * ctl->unit, 1);
2409 * This searches the block group for just extents to fill the cluster with.
2410 * Try to find a cluster with at least bytes total bytes, at least one
2411 * extent of cont1_bytes, and other clusters of at least min_bytes.
2414 setup_cluster_no_bitmap(struct btrfs_block_group_cache *block_group,
2415 struct btrfs_free_cluster *cluster,
2416 struct list_head *bitmaps, u64 offset, u64 bytes,
2417 u64 cont1_bytes, u64 min_bytes)
2419 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2420 struct btrfs_free_space *first = NULL;
2421 struct btrfs_free_space *entry = NULL;
2422 struct btrfs_free_space *last;
2423 struct rb_node *node;
2429 entry = tree_search_offset(ctl, offset, 0, 1);
2434 * We don't want bitmaps, so just move along until we find a normal
2437 while (entry->bitmap || entry->bytes < min_bytes) {
2438 if (entry->bitmap && list_empty(&entry->list))
2439 list_add_tail(&entry->list, bitmaps);
2440 node = rb_next(&entry->offset_index);
2443 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2446 window_start = entry->offset;
2447 window_free = entry->bytes;
2448 max_extent = entry->bytes;
2452 for (node = rb_next(&entry->offset_index); node;
2453 node = rb_next(&entry->offset_index)) {
2454 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2456 if (entry->bitmap) {
2457 if (list_empty(&entry->list))
2458 list_add_tail(&entry->list, bitmaps);
2462 if (entry->bytes < min_bytes)
2466 window_free += entry->bytes;
2467 if (entry->bytes > max_extent)
2468 max_extent = entry->bytes;
2471 if (window_free < bytes || max_extent < cont1_bytes)
2474 cluster->window_start = first->offset;
2476 node = &first->offset_index;
2479 * now we've found our entries, pull them out of the free space
2480 * cache and put them into the cluster rbtree
2485 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2486 node = rb_next(&entry->offset_index);
2487 if (entry->bitmap || entry->bytes < min_bytes)
2490 rb_erase(&entry->offset_index, &ctl->free_space_offset);
2491 ret = tree_insert_offset(&cluster->root, entry->offset,
2492 &entry->offset_index, 0);
2493 total_size += entry->bytes;
2494 ASSERT(!ret); /* -EEXIST; Logic error */
2495 } while (node && entry != last);
2497 cluster->max_size = max_extent;
2498 trace_btrfs_setup_cluster(block_group, cluster, total_size, 0);
2503 * This specifically looks for bitmaps that may work in the cluster, we assume
2504 * that we have already failed to find extents that will work.
2507 setup_cluster_bitmap(struct btrfs_block_group_cache *block_group,
2508 struct btrfs_free_cluster *cluster,
2509 struct list_head *bitmaps, u64 offset, u64 bytes,
2510 u64 cont1_bytes, u64 min_bytes)
2512 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2513 struct btrfs_free_space *entry;
2515 u64 bitmap_offset = offset_to_bitmap(ctl, offset);
2517 if (ctl->total_bitmaps == 0)
2521 * The bitmap that covers offset won't be in the list unless offset
2522 * is just its start offset.
2524 entry = list_first_entry(bitmaps, struct btrfs_free_space, list);
2525 if (entry->offset != bitmap_offset) {
2526 entry = tree_search_offset(ctl, bitmap_offset, 1, 0);
2527 if (entry && list_empty(&entry->list))
2528 list_add(&entry->list, bitmaps);
2531 list_for_each_entry(entry, bitmaps, list) {
2532 if (entry->bytes < bytes)
2534 ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset,
2535 bytes, cont1_bytes, min_bytes);
2541 * The bitmaps list has all the bitmaps that record free space
2542 * starting after offset, so no more search is required.
2548 * here we try to find a cluster of blocks in a block group. The goal
2549 * is to find at least bytes+empty_size.
2550 * We might not find them all in one contiguous area.
2552 * returns zero and sets up cluster if things worked out, otherwise
2553 * it returns -enospc
2555 int btrfs_find_space_cluster(struct btrfs_root *root,
2556 struct btrfs_block_group_cache *block_group,
2557 struct btrfs_free_cluster *cluster,
2558 u64 offset, u64 bytes, u64 empty_size)
2560 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2561 struct btrfs_free_space *entry, *tmp;
2568 * Choose the minimum extent size we'll require for this
2569 * cluster. For SSD_SPREAD, don't allow any fragmentation.
2570 * For metadata, allow allocates with smaller extents. For
2571 * data, keep it dense.
2573 if (btrfs_test_opt(root, SSD_SPREAD)) {
2574 cont1_bytes = min_bytes = bytes + empty_size;
2575 } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) {
2576 cont1_bytes = bytes;
2577 min_bytes = block_group->sectorsize;
2579 cont1_bytes = max(bytes, (bytes + empty_size) >> 2);
2580 min_bytes = block_group->sectorsize;
2583 spin_lock(&ctl->tree_lock);
2586 * If we know we don't have enough space to make a cluster don't even
2587 * bother doing all the work to try and find one.
2589 if (ctl->free_space < bytes) {
2590 spin_unlock(&ctl->tree_lock);
2594 spin_lock(&cluster->lock);
2596 /* someone already found a cluster, hooray */
2597 if (cluster->block_group) {
2602 trace_btrfs_find_cluster(block_group, offset, bytes, empty_size,
2605 INIT_LIST_HEAD(&bitmaps);
2606 ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset,
2608 cont1_bytes, min_bytes);
2610 ret = setup_cluster_bitmap(block_group, cluster, &bitmaps,
2611 offset, bytes + empty_size,
2612 cont1_bytes, min_bytes);
2614 /* Clear our temporary list */
2615 list_for_each_entry_safe(entry, tmp, &bitmaps, list)
2616 list_del_init(&entry->list);
2619 atomic_inc(&block_group->count);
2620 list_add_tail(&cluster->block_group_list,
2621 &block_group->cluster_list);
2622 cluster->block_group = block_group;
2624 trace_btrfs_failed_cluster_setup(block_group);
2627 spin_unlock(&cluster->lock);
2628 spin_unlock(&ctl->tree_lock);
2634 * simple code to zero out a cluster
2636 void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster)
2638 spin_lock_init(&cluster->lock);
2639 spin_lock_init(&cluster->refill_lock);
2640 cluster->root = RB_ROOT;
2641 cluster->max_size = 0;
2642 INIT_LIST_HEAD(&cluster->block_group_list);
2643 cluster->block_group = NULL;
2646 static int do_trimming(struct btrfs_block_group_cache *block_group,
2647 u64 *total_trimmed, u64 start, u64 bytes,
2648 u64 reserved_start, u64 reserved_bytes)
2650 struct btrfs_space_info *space_info = block_group->space_info;
2651 struct btrfs_fs_info *fs_info = block_group->fs_info;
2656 spin_lock(&space_info->lock);
2657 spin_lock(&block_group->lock);
2658 if (!block_group->ro) {
2659 block_group->reserved += reserved_bytes;
2660 space_info->bytes_reserved += reserved_bytes;
2663 spin_unlock(&block_group->lock);
2664 spin_unlock(&space_info->lock);
2666 ret = btrfs_error_discard_extent(fs_info->extent_root,
2667 start, bytes, &trimmed);
2669 *total_trimmed += trimmed;
2671 btrfs_add_free_space(block_group, reserved_start, reserved_bytes);
2674 spin_lock(&space_info->lock);
2675 spin_lock(&block_group->lock);
2676 if (block_group->ro)
2677 space_info->bytes_readonly += reserved_bytes;
2678 block_group->reserved -= reserved_bytes;
2679 space_info->bytes_reserved -= reserved_bytes;
2680 spin_unlock(&space_info->lock);
2681 spin_unlock(&block_group->lock);
2687 static int trim_no_bitmap(struct btrfs_block_group_cache *block_group,
2688 u64 *total_trimmed, u64 start, u64 end, u64 minlen)
2690 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2691 struct btrfs_free_space *entry;
2692 struct rb_node *node;
2698 while (start < end) {
2699 spin_lock(&ctl->tree_lock);
2701 if (ctl->free_space < minlen) {
2702 spin_unlock(&ctl->tree_lock);
2706 entry = tree_search_offset(ctl, start, 0, 1);
2708 spin_unlock(&ctl->tree_lock);
2713 while (entry->bitmap) {
2714 node = rb_next(&entry->offset_index);
2716 spin_unlock(&ctl->tree_lock);
2719 entry = rb_entry(node, struct btrfs_free_space,
2723 if (entry->offset >= end) {
2724 spin_unlock(&ctl->tree_lock);
2728 extent_start = entry->offset;
2729 extent_bytes = entry->bytes;
2730 start = max(start, extent_start);
2731 bytes = min(extent_start + extent_bytes, end) - start;
2732 if (bytes < minlen) {
2733 spin_unlock(&ctl->tree_lock);
2737 unlink_free_space(ctl, entry);
2738 kmem_cache_free(btrfs_free_space_cachep, entry);
2740 spin_unlock(&ctl->tree_lock);
2742 ret = do_trimming(block_group, total_trimmed, start, bytes,
2743 extent_start, extent_bytes);
2749 if (fatal_signal_pending(current)) {
2760 static int trim_bitmaps(struct btrfs_block_group_cache *block_group,
2761 u64 *total_trimmed, u64 start, u64 end, u64 minlen)
2763 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2764 struct btrfs_free_space *entry;
2768 u64 offset = offset_to_bitmap(ctl, start);
2770 while (offset < end) {
2771 bool next_bitmap = false;
2773 spin_lock(&ctl->tree_lock);
2775 if (ctl->free_space < minlen) {
2776 spin_unlock(&ctl->tree_lock);
2780 entry = tree_search_offset(ctl, offset, 1, 0);
2782 spin_unlock(&ctl->tree_lock);
2788 ret2 = search_bitmap(ctl, entry, &start, &bytes);
2789 if (ret2 || start >= end) {
2790 spin_unlock(&ctl->tree_lock);
2795 bytes = min(bytes, end - start);
2796 if (bytes < minlen) {
2797 spin_unlock(&ctl->tree_lock);
2801 bitmap_clear_bits(ctl, entry, start, bytes);
2802 if (entry->bytes == 0)
2803 free_bitmap(ctl, entry);
2805 spin_unlock(&ctl->tree_lock);
2807 ret = do_trimming(block_group, total_trimmed, start, bytes,
2813 offset += BITS_PER_BITMAP * ctl->unit;
2816 if (start >= offset + BITS_PER_BITMAP * ctl->unit)
2817 offset += BITS_PER_BITMAP * ctl->unit;
2820 if (fatal_signal_pending(current)) {
2831 int btrfs_trim_block_group(struct btrfs_block_group_cache *block_group,
2832 u64 *trimmed, u64 start, u64 end, u64 minlen)
2838 ret = trim_no_bitmap(block_group, trimmed, start, end, minlen);
2842 ret = trim_bitmaps(block_group, trimmed, start, end, minlen);
2848 * Find the left-most item in the cache tree, and then return the
2849 * smallest inode number in the item.
2851 * Note: the returned inode number may not be the smallest one in
2852 * the tree, if the left-most item is a bitmap.
2854 u64 btrfs_find_ino_for_alloc(struct btrfs_root *fs_root)
2856 struct btrfs_free_space_ctl *ctl = fs_root->free_ino_ctl;
2857 struct btrfs_free_space *entry = NULL;
2860 spin_lock(&ctl->tree_lock);
2862 if (RB_EMPTY_ROOT(&ctl->free_space_offset))
2865 entry = rb_entry(rb_first(&ctl->free_space_offset),
2866 struct btrfs_free_space, offset_index);
2868 if (!entry->bitmap) {
2869 ino = entry->offset;
2871 unlink_free_space(ctl, entry);
2875 kmem_cache_free(btrfs_free_space_cachep, entry);
2877 link_free_space(ctl, entry);
2883 ret = search_bitmap(ctl, entry, &offset, &count);
2884 /* Logic error; Should be empty if it can't find anything */
2888 bitmap_clear_bits(ctl, entry, offset, 1);
2889 if (entry->bytes == 0)
2890 free_bitmap(ctl, entry);
2893 spin_unlock(&ctl->tree_lock);
2898 struct inode *lookup_free_ino_inode(struct btrfs_root *root,
2899 struct btrfs_path *path)
2901 struct inode *inode = NULL;
2903 spin_lock(&root->cache_lock);
2904 if (root->cache_inode)
2905 inode = igrab(root->cache_inode);
2906 spin_unlock(&root->cache_lock);
2910 inode = __lookup_free_space_inode(root, path, 0);
2914 spin_lock(&root->cache_lock);
2915 if (!btrfs_fs_closing(root->fs_info))
2916 root->cache_inode = igrab(inode);
2917 spin_unlock(&root->cache_lock);
2922 int create_free_ino_inode(struct btrfs_root *root,
2923 struct btrfs_trans_handle *trans,
2924 struct btrfs_path *path)
2926 return __create_free_space_inode(root, trans, path,
2927 BTRFS_FREE_INO_OBJECTID, 0);
2930 int load_free_ino_cache(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2932 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
2933 struct btrfs_path *path;
2934 struct inode *inode;
2936 u64 root_gen = btrfs_root_generation(&root->root_item);
2938 if (!btrfs_test_opt(root, INODE_MAP_CACHE))
2942 * If we're unmounting then just return, since this does a search on the
2943 * normal root and not the commit root and we could deadlock.
2945 if (btrfs_fs_closing(fs_info))
2948 path = btrfs_alloc_path();
2952 inode = lookup_free_ino_inode(root, path);
2956 if (root_gen != BTRFS_I(inode)->generation)
2959 ret = __load_free_space_cache(root, inode, ctl, path, 0);
2963 "failed to load free ino cache for root %llu",
2964 root->root_key.objectid);
2968 btrfs_free_path(path);
2972 int btrfs_write_out_ino_cache(struct btrfs_root *root,
2973 struct btrfs_trans_handle *trans,
2974 struct btrfs_path *path,
2975 struct inode *inode)
2977 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
2980 if (!btrfs_test_opt(root, INODE_MAP_CACHE))
2983 ret = __btrfs_write_out_cache(root, inode, ctl, NULL, trans, path, 0);
2985 btrfs_delalloc_release_metadata(inode, inode->i_size);
2987 btrfs_err(root->fs_info,
2988 "failed to write free ino cache for root %llu",
2989 root->root_key.objectid);
2996 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
2998 * Use this if you need to make a bitmap or extent entry specifically, it
2999 * doesn't do any of the merging that add_free_space does, this acts a lot like
3000 * how the free space cache loading stuff works, so you can get really weird
3003 int test_add_free_space_entry(struct btrfs_block_group_cache *cache,
3004 u64 offset, u64 bytes, bool bitmap)
3006 struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
3007 struct btrfs_free_space *info = NULL, *bitmap_info;
3014 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
3020 spin_lock(&ctl->tree_lock);
3021 info->offset = offset;
3022 info->bytes = bytes;
3023 ret = link_free_space(ctl, info);
3024 spin_unlock(&ctl->tree_lock);
3026 kmem_cache_free(btrfs_free_space_cachep, info);
3031 map = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
3033 kmem_cache_free(btrfs_free_space_cachep, info);
3038 spin_lock(&ctl->tree_lock);
3039 bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
3044 add_new_bitmap(ctl, info, offset);
3048 bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes);
3049 bytes -= bytes_added;
3050 offset += bytes_added;
3051 spin_unlock(&ctl->tree_lock);
3062 * Checks to see if the given range is in the free space cache. This is really
3063 * just used to check the absence of space, so if there is free space in the
3064 * range at all we will return 1.
3066 int test_check_exists(struct btrfs_block_group_cache *cache,
3067 u64 offset, u64 bytes)
3069 struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
3070 struct btrfs_free_space *info;
3073 spin_lock(&ctl->tree_lock);
3074 info = tree_search_offset(ctl, offset, 0, 0);
3076 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
3084 u64 bit_off, bit_bytes;
3086 struct btrfs_free_space *tmp;
3089 bit_bytes = ctl->unit;
3090 ret = search_bitmap(ctl, info, &bit_off, &bit_bytes);
3092 if (bit_off == offset) {
3095 } else if (bit_off > offset &&
3096 offset + bytes > bit_off) {
3102 n = rb_prev(&info->offset_index);
3104 tmp = rb_entry(n, struct btrfs_free_space,
3106 if (tmp->offset + tmp->bytes < offset)
3108 if (offset + bytes < tmp->offset) {
3109 n = rb_prev(&info->offset_index);
3116 n = rb_next(&info->offset_index);
3118 tmp = rb_entry(n, struct btrfs_free_space,
3120 if (offset + bytes < tmp->offset)
3122 if (tmp->offset + tmp->bytes < offset) {
3123 n = rb_next(&info->offset_index);
3133 if (info->offset == offset) {
3138 if (offset > info->offset && offset < info->offset + info->bytes)
3141 spin_unlock(&ctl->tree_lock);
3144 #endif /* CONFIG_BTRFS_FS_RUN_SANITY_TESTS */